Improving effect size estimation and statistical power with multi-echo fMRI and its impact on understanding the neural systems supporting mentalizing

Michael V Lombardo,Bonnie Auyeung,Rosemary J Holt,Jack Waldman,Amber N.V Ruigrok,Natasha Mooney,Edward T Bullmore,Simon Baron-Cohen,Prantik Kundu

doi:10.1016/j.neuroimage.2016.07.022

Abstract

Functional magnetic resonance imaging (fMRI) research is routinely criticized for being statistically underpowered due to characteristically small sample sizes and much larger sample sizes are being increasingly recommended. Additionally, various sources of artifact inherent in fMRI data can have detrimental impact on effect size estimates and statistical power. Here we show how specific removal of non-BOLD artifacts can improve effect size estimation and statistical power in task-fMRI contexts, with particular application to the social-cognitive domain of mentalizing/theory of mind. Non-BOLD variability identification and removal is achieved in a biophysical and statistically principled manner by combining multi-echo fMRI acquisition and independent components analysis (ME-ICA). Without smoothing, group-level effect size estimates on two different mentalizing tasks were enhanced by ME-ICA at a median rate of 24% in regions canonically associated with mentalizing, while much more substantial boosts (40–149%) were observed in non-canonical cerebellar areas. Effect size boosting occurs via reduction of non-BOLD noise at the subject-level and consequent reductions in between-subject variance at the group-level. Smoothing can attenuate ME-ICA-related effect size improvements in certain circumstances. Power simulations demonstrate that ME-ICA-related effect size enhancements enable much higher-powered studies at traditional sample sizes. Cerebellar effects observed after applying ME-ICA may be unobservable with conventional imaging at traditional sample sizes. Thus, ME-ICA allows for principled design-agnostic non-BOLD artifact removal that can substantially improve effect size estimates and statistical power in task-fMRI contexts. ME-ICA could mitigate some issues regarding statistical power in fMRI studies and enable novel discovery of aspects of brain organization that are currently under-appreciated and not well understood.

Highlights

A common criticism of neuroscience research in general (Button et al, 2013) and functional MRI in particular (Yarkoni, 2009), is that studies are characteristically statistically underpowered
ME data show the signal evolution of susceptibility artifact in areas such as ventromedial prefrontal cortex — it is made clear from Fig. 1A that signal dropout occurs at longer TEs, as affected regions have short T2* due to proximity to air-tissue boundaries
Typical task-based functional MRI (fMRI) studies do not apply advanced methods to mitigate substantial non-blood oxygenation level dependent (BOLD) noise that is generally known to be inherent in such data

Summary

Introduction

A common criticism of neuroscience research in general (Button et al, 2013) and functional MRI (fMRI) in particular (Yarkoni, 2009), is that studies are characteristically statistically underpowered. Low statistical power by definition means that a study will have less of a chance for detecting true effects, and means that observed statistically significant effects are less likely to be true and will be more susceptible to the biasing impact of questionable research practices (Button et al, 2013; Ioannidis, 2005). This problem is important given the emergent ‘crisis of confidence’ across many domains of science (e.g., psychology, neuroscience), stemming from low frequency of replication and the pervasive nature of questionable research practices (Button et al, 2013; Ioannidis, 2005; Simmons et al, 2011). These considerations are especially poignant when mandates for large-N studies and increased within-subject scan time are practically limiting due to often cited reasons such as prohibitively high imaging costs for all but the most well-funded research groups or in situations where the focus is on studying sensitive, rare, and/or less prevalent patient populations and where increasing scan time is impractical (e.g., children, patients with neuropsychiatric conditions)

Methods

Results

Conclusion